Skip to main content
computer-science transportation

Multi hop routing

Description

A pattern where items cannot travel direct from source to destination but instead transit intermediate hubs, each of which knows only how to forward to the next hop — not where the cargo ultimately ends up. The routing topology is a structural primitive distinct from both the cargo and any individual hop: the same topology routes anything, and the same cargo can flow through different topologies. The diagnostic shape: source → A → B → C → destination, where A doesn’t know about destination; A only knows about B. The concept is structurally distinct from loop-completion (which is about closure of a route through hubs — the diagnostic question is “does the route close?”), from flow (which is about directed substance-movement — the diagnostic question is “what flows?”), and from asymmetric-gate (which is about boundary checks at hops — the diagnostic question is “can crossing happen?”). Multi-hop-routing’s diagnostic question is “what does each hop know, and what does it not need to know?”

Triggers

User-initiated: User describes indirect routing, hops, intermediate forwarding, transit, or “X goes to Y via Z.” Vocabulary cues: “multi-hop,” “transit,” “relay,” “forwarding,” “next hop,” “hub,” “broker,” “gateway.” Agent-initiated: Agent notices a system where transit happens via intermediate stages that don’t know the full path. Candidate inference: “what’s the per-hop information; what does each hub know about the topology?” Situation-shape signals: Systems where you can’t directly connect source to destination. Conversations about routing protocols, transit, or “how does X get to Y from here.” Architectural discussions involving brokers, gateways, or message buses.

Exclusions

  • Direct point-to-point — when source and destination are directly connected, there are no intermediate hubs. The concept doesn’t fire.
  • Broadcast / flood — when “routing” means “everyone gets it,” there’s no per-hop next-hop decision; the topology is the whole graph.
  • Routes encoded in the cargo itself — source-routed packets (IP source routing; some carrier-pigeon scenarios) put the full path in the message; multi-hop-routing’s defining property is per-hop local-only decisions.
  • Trivially-known topology — if every hop knows the full destination explicitly, the concept collapses to “directed graph traversal” without the load-bearing per-hop-locality insight.

Structure

Internal structure of multi-hop-routing: a table of its component slots and the concepts that fill them.

Relationships

Relationship neighborhood of multi-hop-routing: a graph of the concepts it connects to and the concepts it is a part of.
  • shape — the routing topology is shape independent of cargo.
  • stack-layer — each hop is a layer; the routing decision often happens at the layer boundary; OSI model frames this explicitly.
  • seam — hop boundaries are seams; the routing decision lives at the seam.
  • flow — multi-hop-routing is flow through a non-direct topology; the topology is what distinguishes it from simple source-to-destination flow.
  • grain — what counts as a single hop is a grain choice; an “L1 hop” might be one BGP-router; an “L2 hop” might be one ASN; the choice changes the analysis.

Examples

Internet packet routing (IP / BGP) · computer-science

packets transit routers via per-hop next-hop tables; no router needs to know the entire path.

Bus-route transfers · transportation

passenger takes bus A to a transfer point, then bus B, then bus C; bus A’s driver doesn’t need to know the final destination.
The internet’s interdomain and intradomain routing infrastructure is built explicitly on per-hop forwarding. BGP (Border Gateway Protocol) handles routing between autonomous systems on the global internet; IS-IS and OSPF handle link-state routing within an autonomous system. In each protocol, no router carries the full path from source to destination — each router holds only a forwarding table that maps destination prefix to next hop, and a packet bounces from router to router until it reaches a network that knows the destination directly. The routing protocols themselves communicate reachability and topology updates among adjacent routers, allowing the global per-hop tables to converge on consistent paths without any single router needing the global view.Inference: The internet’s routing architecture is the canonical worked example of multi-hop-routing’s load-bearing property — per-hop local-only decisions enable global reachability without global state. The same structural primitive recurs wherever scale, autonomy, or fault-tolerance forbid centralized routing knowledge: peer-to-peer overlay networks, onion-routing systems like Tor, mesh networks, and gossip protocols. Recognizing the per-hop-locality requirement also sharpens the exclusion: protocols that put the full route into the packet (source-routed IP, label-switched MPLS at the LSP level) are not multi-hop-routing in the catalog’s sense, even though their packets traverse multiple hops.
The two standard distributed-systems references treat multi-hop routing from complementary angles. Lynch’s Distributed Algorithms (1996) gives the formal-algorithmic account: nodes are I/O automata, links are channels, and the question is the correctness and message complexity of distributed shortest-path algorithms — asynchronous Bellman-Ford (the theoretical basis of distance-vector routing) and spanning-tree construction that guarantee a message eventually reaches its destination without loops. Tanenbaum’s Computer Networks gives the architectural account: it draws the explicit line between forwarding (a router looks up the destination and sends the packet out the right line) and routing (the process that builds and updates those tables), and describes store-and-forward switching where each intermediate node receives, buffers, and relays the packet onward.Inference: The pair pins down the concept’s roles from both ends. The source and destination are endpoints with no direct link; the intermediate nodes are the relaying routers/automata; the per-hop forwarding decision is the local table lookup that, composed across hops, produces a global path no single node planned in full. The structural lesson both texts converge on is that multi-hop routing is locally decided, globally emergent: no node holds the whole route, yet correct local forwarding rules (and the distributed algorithms that maintain them) suffice to deliver end-to-end — which is exactly what makes the shape scale and survive partial failure.
each peer knows only a local subset; messages propagate via repeated next-hop forwarding.
cross-domain field-validated; the structural primitive is well-established across multiple distinct engineering traditions
virtual addresses route physical mail through chains the sender doesn’t see.
Kafka topics with exchange-bindings; broker forwards without knowing the consumer’s identity.
RFC 791 specifies IPv4: packets carry source and destination addresses but no end-to-end route. Each router along the way inspects the destination, consults its local routing table, and forwards the packet to the next hop only — no router knows the full path the packet will take, and the path may differ for successive packets between the same endpoints. The packet traverses an unknown number of intermediate routers before arriving.Inference: The “per-hop knowledge, no end-to-end knowledge” property is the structural primitive. The same shape recurs in mail forwarding, supply chains with intermediate distributors, gossip protocols, BGP/OSPF inter-router announcements, Tor onion-routing, and graph-traversal algorithms. The diagnostic for a candidate multi-hop-routing match is “do intermediate participants know only the next hop, not the full path?” — distinct from end-to-end routing (where the sender knows the path) and from broadcast (where the destination isn’t named at all).
manufacturer → wholesaler → retailer → consumer; each stage’s bookkeeping covers only its adjacent neighbors.
explicit anonymity application: each hop knows only the previous and next hops, not the full circuit.